Object detection models can be trained to identify predefined objects from digital photographs with a certain accuracy level. This model uses Artificial Intelligence (AI) to analyze the images that you upload. 

You may have a huge inventory of diverse objects that require manual detection on a daily basis, which might be time-consuming and repetitive. This is where the Object Detection model comes in to automate the process. It identifies objects by comparing the uploaded user-input data to the predefined objects that the model can recognize. Using this model, one can automate business processes like maintaining an inventory, accounting, and more.

Object Detection model prerequisites

Businesses of any size have their own custom objects to be detected. For example, if you're dealing with construction materials, you can upload images of various bricks used and train the model to identify each of them.

Creator enables you to build and train custom object detection models tailored to suit your business needs. 

You can also utilize our ready-to-use models to detect objects from a pre-defined list of objects. In ready-to-use models, you can select the application name and form name from the Microservices section and you'll be taken to the form in which a new object detection field will be created. 

What skills do you need?

• As the admin, you can create and use object detection models, whereas your users only consume the models you've created. 
• You should know enough about all your business requirements to determine and gather the images that'll be used for training the model.
• As a low-code platform, Creator doesn't require you and your users to have prior coding and machine-learning skills to create and consume the object detection models.

What data do you need?

• You need a minimum of 10 images for each object to train an object detection model. For desirable results, you should have at least 20 images.
• You can also use 50 images to help the model identify image attributes (color, shape, size) better and also improve the model performance. 
• For desirable results, you can upload up to 20 images in varying angles.
• For model training, you may upload 10 images for each object, though more is preferable. The model would be much more confident and be able to recognize image properties such as color, shape, and size in 50 images.

Which pricing plan must you be in?

• AI models will be available for users in Paid Plans of Creator. Refer to our pricing page to learn more.

Which version of Creator should you be using?

• You must be using Zoho Creator 6 (C6) to be able to create custom AI models whereas, the ready-to-use AI models are available in both C6 and C5.

Get started with sample data

To help you get started quickly and explore the possibilities of object detection, you can build and train an object detection model using sample pictures, that you can readily use in your Creator applications. 

Let's assume that you've built Zylker's Inventory Management app using Creator, and you want to detect various objects to keep an account of the number of items in stock. Manually accounting for a large-scale inventory is very time-consuming. 

Let's say this app keeps count of three construction raw material: hollow concrete blocks, AAC blocks, and lintel blocks.

Here's how you would use the Object Detection model in this case:

1. Create a model:

• Identify the different type of objects that you want to account for and be identified by the model. Here, the objects that require to be identified include hollow concrete blocks, AAC blocks, and lintel blocks.
• Upload sufficient images of the objects that need to be identified. Here, you'll need to upload images of the three types of bricks.
• Make sure to upload a minimum of 10 images for each object, preferably with a plain background during training. Ensure that you cover the object in different angles and lighting.

2. Train the model.
3. Deploy the model: 

• Select an image field in the form which will contain the input for the model. 
• Add a new object detection field.
• Upload an image in live to get the required values. The input in this case will be the image of one among the three bricks.

Object Detection Model Flow

Add Training Data

Training data is the initial dataset that is used by the model to analyze data patterns, make interpretations, and arrive at a learning that helps it make detections. To train an object detection model to recognize your objects, you have to gather sufficient images that contain those objects. 

Format and size

The images that you'll be feeding your object detection model require the following:

• Format:

• JPG
• PNG
• TIF

• Size:

• Maximum 5 MB per image in an object
• Overall model size is 250 MB

Guidelines

• For an object detection model to function, a minimum of two objects need to be added as training data. 
• We recommend that you upload at least 10 images each for the objects that you add.
• Ensure that the overall model size does not exceed 250 MB.
• If you upload a folder containing all the required images, we import only the first 20 images from that folder.
• If the size of the image folder exceeds the model size limit of 250 MB, only a few images will be uploaded, provided their size is within the allowed model size.
• To add an object detection field in your form, you must have an image field as the source field in that form.
• If none of the supported field types are available in the form, you will first need to create them in order to deploy the object detection model. 

Adding images

You need to gather sufficient images that contain the objects you want the model to detect. After collecting images, you need to create an object folder and specify a name for it. You can then upload the required images one by one or all at once. You can also upload the images stored in your .zip folder. In this case, the .zip folder's name will be applied as the newly-created object folder's name. The images that you collected earlier will now come in handy because you need to upload them to the object detection model.

Model Summary

After adding the training data, you can review the model details such as model name,  its type, size and total number of images added per model. You can also view the object details such as object name, trained images, and object size. In case you need to make any modifications, you can go back and make them. Otherwise, you can proceed to train the model. 

• Train model 
• View and manage model details
• Test model
• Publish and use model

Train model

Before you can actually use your object detection model in your application, you have to train it to perform and produce the expected outcomes as per your business needs. After you've selected and reviewed your object images, click Train to train your model.

View and manage model details

Once training your model is complete, you can view the model details, the model's versions and its deployment details, if any. The model is then ready to be published and deployed into your apps.

You can manage your model in the following ways:

• Retrain - This helps in improving the object detection model's reliability and accuracy. 

• You can click Re-train and your model will be retrained.
• After each retraining is over, a new version of the model will be created. You can switch between different versions according to your needs.
• If you want to delete a version that is currently used, you need to switch to another version before proceeding to delete it. 
• If the model training fails, you'll see 'Model training has failed!' Meanwhile, the previous working model will be used for prediction. 

• Rename - You can rename your model if required.
• Delete - If you want to delete your model due to inconsistent or wrongly-added data, you can use the Delete option.

Test model

After training your model, you can test your model to check and know how your model works and if the training is satisfactory, before publishing and deploying it in your applications. You can upload a suitable test data (object image) and after testing your model, you'll get the predicted outcome along with a confidence score. 

• If the confidence score is high (85 - 97), you can proceed to publish your model.
• If the confidence score is good (70 - 84) or fair (51 - 69), you can re-train your model with additional images. 
• If the confidence score is poor(< 50), you can check for inconsistencies in your data, refine them and train your model again.

Confidence score

The object detection model calculates the confidence score for your trained model based on the detection results of your test dataset. For example: if you've uploaded a image as test data that is similar to your training images, you'll get a higher confidence score of 96. If you get a lower confidence score, refer here to improve your model performance.

Improve model performance

• Data quantity - It's recommended to upload sufficient (at least 10) images to train your object detection model and increase its accuracy. With fewer images, there's a strong chance that your model might detect incorrectly or with a lesser confidence score. Also, your model won't be biased and will be able to generalize well if you include various types of images of the same object.
• Data quality - It is recommended to add clear images with a plain well-lit background as training data. With a background, there can be other objects that the model might pick up.
• Data size and angles - Upload images of objects in various sizes that show various aspects of the objects, such as a photo of a group of apples and a close-up of a single apple. This type of difference in scaling aids in the model's ability to generalize objects and identify them with better confidence. It is  also recommended to provide images of objects taken from different angles. 

Publish and use model

After you train your object detection model, you can publish it to make it available to your users. Once the model is deployed in your applications, a new object detection field will be created and you can start detecting objects. Learn how

To use your object detection model in an environment-enabled Creator application, you must have at least one version of that application published in the production environment. After deploying your model in the application, you can filter between different stages of the environment to check which stage the model is deployed in. Learn how

Related Topics

• Understand AI Modeler
• Build and manage object detection model
• Ready-to-use object detection model